Hospitalizations and costs of ambulatory care sensitive conditions in Brazil

Exploratory and Data Analysis

Introduction

About the data

Ambulatory care sensitive conditions refer to health issues that can be effectively prevented, detected early, treated, and managed through primary health care (APS)¹. In Brazil, the Ministry of Health developed its own list of hospitalizations related to these conditions, known as Hospitalizations for Primary Care Sensitive Conditions (ICSAP)^2,3.High rates of ICSAP admissions often indicate serious issues with health system access or performance, reflecting potential gaps in preventive care, early diagnosis, or continuity of treatment¹. Beyond the clinical implications, these hospitalizations carry a significant economic burden, with avoidable admissions representing billions in health care expenditures.

Objective

This exploratory data analysis aims to quantify the clinical and financial impact of ICSAP in the Brazilian SUS from 2014 to 2024, while assessing data quality and identifying inconsistencies to ensure robust insights on primary care gaps and potential cost savings.

Methodology

Hospital Information System (SIH), is an administrative database managed by the Ministry of Health, which also provides the official ICSAP list³. Macro-region data are sourced from the Brazilian Institute of Geography and Statistics (IBGE).

When processing the SIH-SUS data into the silver layer, hospitalizations whose primary diagnosis was related to obstetric causes or childbirth (ICD-10 codes O80–O84) were excluded⁴.

This document presents descriptive analyses of the percentage of ICSAP admissions and their costs, broken down annually by age group, gender, and Brazilian macro-regions. Results are initially presented for the period 2014–2024, followed by a more detailed look at data from 2024. Additionally, the most frequent ICSAP admissions are analyzed according to their primary diagnoses.

It is important to note that the values presented here may differ from other publications due to specific methodological choices made in this study

Data contains

The database consists of multiple datasets. Here will be explored data from the sih_sus table.

1. year: Year of hospitalization (yyyy).
2. state_code: State code of healthcare provider.
3. length_stay: Number of days spent in hospital.
4. gender: Patient’s gender.
5. age: Patient’s age (years).
6. disease_code: Code of the principal diagnostic based in the ICD.10.
7. amount_paid: Costs of the hospital services (BR R$).
8. total_paid: Total costs of the hospitalization (BR R$).
9. us_total_paid: Total costs of the hospitalization (US dolars).

Getting started

The first step is to import all the libraries required for the analysis.

import duckdb
from pathlib import Path
from IPython.display import display
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import plotly.express as px

Next, we establish a connection with the database, as the data is stored in a DuckDB file.

# Set the file path
base_dir = Path.cwd().parent.parent
duckdb_file = base_dir / "sih_sus.duckdb"

# Connect with the database
con = duckdb.connect(str(duckdb_file))

Then, inspect the dimmension of the sih_sus dataset, reviewing the first ten rows, column names and types, basic calculations and presence of null vales.

sih_10 = con.sql("""
SELECT * FROM silver.sih_sus LIMIT 10
""").df()

sih_info = con.sql("""
DESCRIBE silver.sih_sus
""").df()

sih_summa = con.sql("""
SUMMARIZE silver.sih_sus
""").df()

print("First ten roles of the SIH SUS dataset:")
display(sih_10)
print("-------------------------------------------------------------")

print("Table atributes:")
display(sih_info)
print("-------------------------------------------------------------")

print("SIH SUS data summarization:")
display(sih_summa.round(2))

First ten roles of the SIH SUS dataset:

	year	state_code	length_stay	gender	age	disease_code	amount_paid	total_paid	us_total_paid
0	2014	35	13	Male	64	I500	1171.43	1260.71	557.83
1	2014	35	17	Female	60	I802	463.46	548.97	242.90
2	2014	35	7	Male	46	C910	522.76	587.90	260.13
3	2014	35	8	Male	39	J690	1519.22	1636.74	724.22
4	2014	35	16	Female	75	I615	1794.23	2126.65	940.99
5	2014	35	1	Female	48	T838	104.25	138.36	61.22
6	2014	35	0	Female	48	N390	257.47	298.21	131.95
7	2014	35	23	Male	30	J690	6554.89	7591.28	3358.97
8	2014	35	7	Female	48	T838	596.52	732.96	324.31
9	2014	35	10	Female	61	A499	793.69	865.91	383.14

-------------------------------------------------------------
Table atributes:

	column_name	column_type	null	key	default	extra
0	year	SMALLINT	YES	None	None	None
1	state_code	SMALLINT	YES	None	None	None
2	length_stay	SMALLINT	YES	None	None	None
3	gender	VARCHAR	YES	None	None	None
4	age	SMALLINT	YES	None	None	None
5	disease_code	VARCHAR	YES	None	None	None
6	amount_paid	DECIMAL(13,2)	YES	None	None	None
7	total_paid	DECIMAL(14,2)	YES	None	None	None
8	us_total_paid	DECIMAL(8,2)	YES	None	None	None

-------------------------------------------------------------
SIH SUS data summarization:

	column_name	column_type	min	max	approx_unique	avg	std	q25	q50	q75	count	null_percentage
0	year	SMALLINT	2014	2024	12	2019.1261591898613	3.204546316556134	2016	2019	2022	134894533	0.0
1	state_code	SMALLINT	11	53	21	32.58773457483262	9.518801432635128	26	33	39	134894533	0.0
2	length_stay	SMALLINT	0	364	356	5.287321525476499	7.7696769384484154	1	3	6	134894533	0.0
3	gender	VARCHAR	Female	Male	2	None	None	None	None	None	134894533	0.0
4	age	SMALLINT	0	99	98	40.98609957751216	24.22237723580786	22	38	61	134894533	0.0
5	disease_code	VARCHAR	A00	Z999	11377	None	None	None	None	None	134894533	0.0
6	amount_paid	DECIMAL(13,2)	0.00	235279.48	1822449	1123.5646298278077	3027.7482491145497	266	424	797	134894533	0.0
7	total_paid	DECIMAL(14,2)	0.00	274207.48	2315519	1414.903087305399	3602.6482743774177	346	602	1042	134894533	0.0
8	us_total_paid	DECIMAL(8,2)	0.00	97950.87	1078372	344.7467544773664	879.947329460279	85	147	263	134894533	0.0

The initial exploration reveals some key characteristcs of the dataset. SIH-SUS data contains 134,894,533 entries. Some categorical variables like year, state_code and age are stored as SMALLINT to optimize storage.

Min and Max values for the year, lenght_stay and age columns suggests that SIH-SUS covers data from 2014 to 2024, that some individual were admitted and discharged at the same day (0 values) and that the data covers a range (0 to 99) of individuals from newborn to elderly. Also, the null_percentage is 0.0 for all columns. The maximum value of 364 days spent shows that there are very long-term hospitalizations in the sih_sus dataset, which is expected but worth verifying.

The Min value of 0 for amount_paid and total_cost is interesting but may not be an error because some scenarios can be recorded as having zero cost, as such:

• If the main payment is registered on the first AIH, subsequent monthly records might appear with a cost of zero.
• When a patient is transferred between hospitals
• When certain procedures or administrative adjustments within the system might not have a direct cost associated with them but still generate a record.

We can count the number of entries with a length of 364 days of stay and with zero cost.

sih_count_364_stay = con.sql("""
SELECT COUNT(*)
FROM silver.sih_sus
WHERE length_stay = 364
""").to_df().iloc[0,0]

sih_count_zero_cost = con.sql("""
SELECT COUNT(*)
FROM silver.sih_sus
WHERE total_paid = 0
""").to_df().iloc[0,0]

print(f"There are {sih_count_364_stay} entries registered with a length of 364 days of stay.")
print(f"There are {sih_count_zero_cost} entries registered with zero cost.")

sih_364_stay = con.sql("""
SELECT *
FROM silver.sih_sus
WHERE length_stay = 364
""").df()

display(sih_364_stay)
print("------------------------------------------------------------------")

sih_zero_cost = con.sql("""
SELECT *
FROM silver.sih_sus
WHERE total_paid = 0
LIMIT 10
""").to_df()

display(sih_zero_cost)

There are 12 entries registered with a length of 364 days of stay.
There are 282216 entries registered with zero cost.

	year	state_code	length_stay	gender	age	disease_code	amount_paid	total_paid	us_total_paid
0	2015	33	364	Male	7	A419	4326.36	4781.71	1666.10
1	2016	35	364	Male	1	Q210	19116.54	22481.91	6917.51
2	2016	33	364	Male	5	N47	97.72	219.12	55.47
3	2017	33	364	Male	9	P271	89222.98	103868.75	31475.37
4	2017	33	364	Male	10	J849	13169.99	14935.84	4526.01
5	2018	33	364	Female	55	K808	447.16	695.77	173.94
6	2019	31	364	Female	61	A418	1885.93	2166.80	580.91
7	2021	33	364	Male	1	Q743	3013.90	3632.71	679.01
8	2021	33	364	Female	1	Q038	7691.41	8762.22	1613.66
9	2022	33	364	Male	2	A415	4980.50	5435.85	1149.22
10	2022	23	364	Male	30	K928	293.57	347.15	70.70
11	2024	11	364	Male	4	G120	6033.36	6453.57	1293.30

------------------------------------------------------------------

	year	state_code	length_stay	gender	age	disease_code	amount_paid	total_paid	us_total_paid
0	2014	35	1	Female	37	F192	0.0	0.0	0.0
1	2014	35	1	Male	26	F200	0.0	0.0	0.0
2	2014	35	1	Male	56	F192	0.0	0.0	0.0
3	2014	35	1	Male	54	F206	0.0	0.0	0.0
4	2014	35	1	Male	38	F102	0.0	0.0	0.0
5	2014	35	1	Male	43	F102	0.0	0.0	0.0
6	2014	35	1	Male	47	F192	0.0	0.0	0.0
7	2014	35	1	Male	33	F192	0.0	0.0	0.0
8	2014	35	1	Male	37	F142	0.0	0.0	0.0
9	2014	35	1	Male	41	F200	0.0	0.0	0.0

It seems that the entries registered with a length of 364 days of stay are valid, since they are distributed across different years, ages and genders.

However, there are many entries registered with zero cost. What can be done about it? For most of analyses that involve calculating averages or sums of costs, it’s a good practice to filter these records out to avoid distorting the results.

Exploratory analysis

The next step is to explore and analyze ICSAP profiles and costs within SIH-SUS by join with the cid10_icsap table.

ICSAP Metrics By Year (2014 to 2024)

How have primary care-sensitive hospitalizations and costs evolved over the last ten years?

# Query the results
# Counting totals
sih_sus_count_year = con.sql("""
SELECT
    sih.year,
    COUNT(*) AS total_hospitalizations,
    COUNT(CASE
              WHEN icsap.avoidable_disease_code IS NOT NULL THEN 1
          END) AS total_icsap,
    ROUND(COUNT(CASE
                    WHEN icsap.avoidable_disease_code IS NOT NULL THEN 1
                END) * 100.0 / COUNT(*), 2) AS rate_icsap
FROM silver.sih_sus AS sih
LEFT JOIN silver.cid10_icsap AS icsap
    ON sih.disease_code = icsap.avoidable_disease_code
WHERE sih.total_paid > 0
GROUP BY sih.year
ORDER BY sih.year
""").to_df()

# Calculating overall costs by year
sih_sus_costs_year = con.sql("""
SELECT
    sih.year,
    SUM(sih.total_paid) AS total_cost,
    SUM(CASE
            WHEN icsap.avoidable_disease_code IS NOT NULL THEN sih.total_paid
        END) AS icsap_cost,
    ROUND(SUM(CASE
                  WHEN icsap.avoidable_disease_code IS NOT NULL THEN sih.total_paid
                  ELSE 0
              END) * 100.0 / SUM(sih.total_paid), 2) AS rate_icsap
FROM silver.sih_sus AS sih
LEFT JOIN silver.cid10_icsap AS icsap
    ON sih.disease_code = icsap.avoidable_disease_code
WHERE sih.total_paid > 0
GROUP BY sih.year
ORDER BY sih.year
""").to_df()

# Create plot
sns.set_theme(style="white", palette="colorblind")
# Bar plot
fig, ax = plt.subplots()

sns.barplot(
    data=sih_sus_count_year,
    x="year", y="rate_icsap",
    ax=ax
)
ax.bar_label(ax.containers[0], fmt='%.1f%%', fontsize=9, color='black', fontweight='bold')
ax.spines[['right', 'top', 'left']].set_visible(False)
ax.yaxis.set_visible(False)
ax.set_xlabel("")
plt.show()

# Currency formatter function
def format_brl(value):
    if value >= 1e9:
        return f"R${value / 1e9:.2f}B"
    elif value >= 1e6:
        return f"R${value / 1e6:.1f}M"
    else:
        return f"R${value:,.0f}"

# Horizontal bar plot
year_order = sorted(sih_sus_count_year["year"].unique(), reverse=True)
df_year_sorted = sih_sus_costs_year.set_index("year").loc[year_order].reset_index()

fig, ax = plt.subplots()

sns.barplot(
    data=df_year_sorted,
    x="icsap_cost", y="year",
    orient='h',
    ax=ax, order=year_order
)

# Format values for bar labels
formatted_values = [format_brl(v) for v in df_year_sorted["icsap_cost"]]

ax.bar_label(ax.containers[0], labels=formatted_values, padding=-65, fontsize=9, color='white', fontweight='bold')
ax.spines[['right', 'top', 'bottom']].set_visible(False)
ax.xaxis.set_visible(False)
ax.set_ylabel("")
plt.show()

(a) Percent of ICSAP Over Total Hospitalizations by year

(b) ICSAP Total Cost by Year

Figure 1: Hospitalizations and Cost by Year

From 2014 to 2024, Brazil experienced a gradual decline in the proportion of avoidable hospitalizations (ICSAP) relative to total hospitalizations, dropping from 9.2% in 2014 to 7.81% in 2024. The absolute number of ICSAP cases remained relatively stable over the years.

In terms of financial impact, ICSAP-related costs followed a similar pattern: while their share of total hospitalization costs fluctuated between 5.25% and 5.79%, the absolute ICSAP cost rose significantly, reaching over R$1.2 billion in 2024.

ICSAP Metrics By year and Brazilian Macro-regions

To add a regional perspective, we join the dataset with the uf_localidade table, which contains information about Brazil’s macro-regions.

sih_sus_regional = con.sql("""
SELECT
    sih.year,
    uf.big_region_name,
    COUNT(*) AS total_hospitalizations,
    COUNT(CASE 
              WHEN cid.avoidable_disease_code IS NOT NULL THEN 1 
          END) AS avoidable_disease_hosp,
    ROUND(COUNT(CASE 
                    WHEN cid.avoidable_disease_code IS NOT NULL THEN 1 
                END) * 100.0 / COUNT(*), 2) AS tx_avoidable_disease,

    ROUND(SUM(sih.total_paid)) AS total_cost,
    ROUND(SUM(CASE 
                  WHEN cid.avoidable_disease_code IS NOT NULL THEN sih.total_paid
                  ELSE 0 
              END), 2) AS avoidable_disease_cost,
    ROUND(SUM(CASE 
                  WHEN cid.avoidable_disease_code IS NOT NULL THEN sih.total_paid
                  ELSE 0 
              END) * 100.0 / SUM(sih.total_paid), 2) AS tx_avoidable_disease_cost 

FROM silver.sih_sus AS sih
LEFT JOIN silver.cid10_icsap AS cid
    ON sih.disease_code = cid.avoidable_disease_code
LEFT JOIN silver.uf_localidade AS uf
    ON sih.state_code = uf.state_code
GROUP BY sih.year, uf.big_region_name
ORDER BY sih.year, uf.big_region_name
""").to_df()

# Create plot
sns.set_theme(style="whitegrid", palette="colorblind")

# Bar plot
fig, ax = plt.subplots(figsize=(10, 6))

sns.barplot(
    data=sih_sus_regional,
    x="year",
    y="tx_avoidable_disease",
    hue="big_region_name", 
    ax=ax
)
ax.spines[['right', 'top']].set_visible(False)
ax.set_ylabel("ICSAP (%)")
ax.set_xlabel("")
ax.legend(title=None, loc="upper center", bbox_to_anchor=(0.5, 1.15), ncol=5)
plt.tight_layout()
plt.show()

# Line Plot
fig, ax = plt.subplots(figsize=(10, 6))
sns.lineplot(
    data=sih_sus_regional,
    x="year", y="tx_avoidable_disease_cost",
    hue="big_region_name",
    marker="o",
    ax=ax
)
ax.set_ylabel("ICSAP Cost (%)")
ax.set_xlabel("")
ax.set_xticks(sorted(sih_sus_regional["year"].unique()))
ax.set_xticklabels(sorted(sih_sus_regional["year"].unique()))
ax.legend(title=None, loc="upper center", bbox_to_anchor=(0.5, 1.15), ncol=5)
plt.tight_layout()
plt.show()

(a) ICSAP (2014 to 2024)

(b) ICSAP Costs (2014 to 2024)

Figure 2: Percent of Avoidable Hospitalizations and Costs by Brazilian Macro-Regions

The North consistently shows the highest ICSAP rates — above 9% annually — indicating structural challenges in primary care access. The Northeast follows closely, with high volumes and cost shares, reflecting its large population and uneven healthcare coverage. Meanwhile, the South and Southeast maintain lower ICSAP rates and cost proportions, though the Southeast still incurs the highest absolute costs due to its large population and hospitalization scale. The Centro-Oeste region remains intermediate in both metrics.

Over time, most regions show a gradual decline in ICSAP rates, yet the financial burden remains substantial, with avoidable costs exceeding R$500 million annually in the Southeast alone.

ICSAP Metrics by the Last Year (2024)

Focusing on 2024, we analyze ICSAP by region, city, age group, and gender.

con.sql("""
SELECT COUNT(*)
FROM silver.sih_sus as sih
LEFT JOIN silver.cid10_icsap AS cid
    ON sih.disease_code = cid.avoidable_disease_code
LEFT JOIN silver.uf_localidade AS uf
    ON sih.state_code = uf.state_code
WHERE sih.year = 2024 
    AND sih.total_paid > 0
    AND cid.avoidable_disease_code IS NOT NULL
""").show()

┌──────────────┐
│ count_star() │
│    int64     │
├──────────────┤
│      1103177 │
└──────────────┘

As the dataset reduced to 1.103,177 rows (about 99%) we will query the data into a Pandas dataframe and continuing with filters and aggregations.

# Query the data
sih_icsap_2024 = con.sql("""
SELECT
    uf.big_region_name,
    uf.state_name,
    sih.length_stay,
    sih.gender,
    sih.age,
    icsap.disease_category,
    sih.total_paid
FROM silver.sih_sus as sih
LEFT JOIN silver.cid10_icsap AS icsap
    ON sih.disease_code = icsap.avoidable_disease_code
LEFT JOIN silver.uf_localidade AS uf
    ON sih.state_code = uf.state_code
WHERE sih.year = 2024
    AND icsap.avoidable_disease_code IS NOT NULL
    AND sih.total_paid > 0
""").to_df()

# Data summary
def column_summary(df):
    summary_data = []

    for col_name in df.columns:
        col_dtype = df[col_name].dtype
        num_of_nulls = df[col_name].isnull().sum()
        num_of_non_nulls = df[col_name].notnull().sum()
        num_of_distinct_values = df[col_name].nunique()

        summary_data.append({
            'col_name': col_name,
            'col_dtype': col_dtype,
            'num_of_nulls': num_of_nulls,
            'num_of_non_nulls': num_of_non_nulls,
            'num_of_distinct_values': num_of_distinct_values
        })
    
    summary_df = pd.DataFrame(summary_data)
    return summary_df

summary_df = column_summary(sih_icsap_2024)

print("Summary of the data:")
display(summary_df)

print("------------------------------------------------------------------------")

describe_df = sih_icsap_2024.describe()
print("Descriptive statistic of the data:")
display(describe_df)

Summary of the data:

	col_name	col_dtype	num_of_nulls	num_of_non_nulls	num_of_distinct_values
0	big_region_name	object	0	1103177	5
1	state_name	object	0	1103177	27
2	length_stay	int16	0	1103177	186
3	gender	object	0	1103177	2
4	age	int16	0	1103177	100
5	disease_category	object	0	1103177	19
6	total_paid	float64	0	1103177	214100

------------------------------------------------------------------------
Descriptive statistic of the data:

	length_stay	age	total_paid
count	1.103177e+06	1.103177e+06	1.103177e+06
mean	6.293990e+00	5.156377e+01	1.129607e+03
std	7.479650e+00	2.800756e+01	2.678277e+03
min	0.000000e+00	0.000000e+00	2.198000e+01
25%	2.000000e+00	2.900000e+01	3.280100e+02
50%	4.000000e+00	5.900000e+01	5.207200e+02
75%	7.000000e+00	7.400000e+01	8.659100e+02
max	3.430000e+02	9.900000e+01	1.679245e+05

Overall, the data reveals high heterogeneity in both cost and clinical profiles, with strong variability across hospitalizations.

The dataset has no missing values. It spans five macro-regions, 27 states, and includes 19 disease categories. The total_paid variable shows high variability, with over 214,000 distinct values and a wide range from R$21.98 to R$167,924.50. The average hospitalization cost is approximately R$1,129.61, but the standard deviation is more than twice that (R$2,678.28), indicating a highly skewed distribution with probably substantial outliers.

Length_stay also exhibits variability, with a mean of 6 days and a standard deviation of 7 days. While most hospitalizations are short (median of 4 days), some extend up to 343 days, suggesting the presence of complex or chronic cases. The average age is 51 years, and the interquartile range (29 to 74) reflects a concentration of adult and elderly patients.

Thus, the statistic suggest that while most cases are relatively short and moderately costly, a subset of patients drives up the average.

So, we can analyse if this could be related due to expanded stays and or age group.

Total Hospitalizations and Costs

First, make a copy of the dataset and create an age group.

# Make a copy of the data
df = sih_icsap_2024.copy()

#Create an age group
bins = [-1, 4, 14, 24, 44, 64, float('inf')]
labels = ['0-4', '5-14', '15-24', '25-44', '45-64', '65+']
df['age_group'] = pd.cut(df['age'], bins=bins, labels=labels, right=True)

df

	big_region_name	state_name	length_stay	gender	age	disease_category	total_paid	age_group
0	Nordeste	Maranhão	14	Male	86	Doenças Cerebrovasculares	575.21	65+
1	Nordeste	Maranhão	4	Female	14	Infecção no Rim e Trato Urinário	275.88	5-14
2	Nordeste	Maranhão	4	Female	1	Gastroenterites Infecciosas e complicações	356.90	0-4
3	Nordeste	Maranhão	4	Male	6	Gastroenterites Infecciosas e complicações	394.85	5-14
4	Nordeste	Maranhão	2	Female	4	Gastroenterites Infecciosas e complicações	340.90	0-4
...	...	...	...	...	...	...	...	...
1103172	Nordeste	Rio Grande do Norte	6	Female	7	Pneumonias bacterianas	630.42	5-14
1103173	Nordeste	Rio Grande do Norte	9	Female	9	Infecção no Rim e Trato Urinário	401.00	5-14
1103174	Nordeste	Rio Grande do Norte	20	Female	2	Deficiências Nutricionais	1949.56	0-4
1103175	Nordeste	Rio Grande do Norte	3	Female	5	Pneumonias bacterianas	697.94	5-14
1103176	Nordeste	Rio Grande do Norte	1	Female	7	Infecção no Rim e Trato Urinário	212.50	5-14

1103177 rows × 8 columns

We can analyse the relationship between lenght of stay and the total cost and visualize the distribution of the age groups by doing a scatter plot.

# Get a sample of the data
sample_df = df.sample(n=100000, random_state=42)

# Create plot
sns.set_theme(style="whitegrid", palette="colorblind")

g = sns.relplot(
    data=sample_df,
    x='length_stay', y='total_paid',
    col='age_group', hue='gender', col_wrap=3,
    kind='scatter', height=4, aspect=1.2, alpha=0.5 
)
sns.move_legend(
    g, "upper center",
    bbox_to_anchor=(.5, 1.0), 
    ncol=2, title=None, frameon=False
)
g.set_axis_labels("Length of Stay (Days)", "Total Cost (R$)")
g.set_titles("Age Group: {col_name}")
plt.tight_layout(rect=[0, 0, 1, 0.97])
plt.show()

# We can filter the data to lenght stays equal or less then 100 days and
# costs equal or less then 60K for better visualization
filtered_df = sample_df[
    (sample_df["length_stay"] <= 100) &
    (sample_df["total_paid"] <= 60000)
]

g = sns.relplot(
    data=filtered_df,
    x='length_stay', y='total_paid',
    col='age_group', hue='gender', col_wrap=3,
    kind='scatter', height=4, aspect=1.2, alpha=0.5 
)
sns.move_legend(
    g, "upper center",
    bbox_to_anchor=(.5, 1.0), 
    ncol=2, title=None, frameon=False
)
g.set_axis_labels("Length of Stay (Days)", "Total Cost (R$)")
g.set_titles("Age Group: {col_name}")
plt.tight_layout(rect=[0, 0, 1, 0.97])
plt.show()

(a) ICSAP Cost vs Length of Stay by Age Group (2024)

(b) ICSAP Cost vs Length of Stay by Age Group (2024) - Filtered

Figure 3: Analysis of the relationship between cost and length of stay

There doesn’t appear to be a clear relationship between length of stay and total cost. We observe both short hospitalizations with high costs and long stays with either high or low costs.

Both age groups are represented across the spectrum, but individuals aged 65+ and 45–64 seem to have a higher concentration of cases with elevated costs and/or extended durations.

The presence of extreme values and wide interquartile ranges suggests that while most hospitalizations are moderately priced, a subset incurs significantly higher costs.

Now, we can analyse the statistic grouped by age group and gender to see the patterms.

summary_gender = df.groupby(["gender"])["total_paid"].describe().round(2)

summary_age = df.groupby(["age_group"])["total_paid"].describe().round(2)

display(summary_gender)
display(summary_age)

	count	mean	std	min	25%	50%	75%	max
gender
Female	557909.0	1055.84	2548.52	21.98	317.64	487.21	805.74	136618.59
Male	545268.0	1205.09	2802.83	21.98	340.90	574.50	900.97	167924.52

	count	mean	std	min	25%	50%	75%	max
age_group
0-4	105421.0	781.47	1858.71	40.38	332.90	447.48	625.44	90287.23
5-14	90342.0	732.19	1633.27	40.38	332.90	410.85	622.42	74917.45
15-24	51335.0	813.00	2089.71	40.38	226.68	332.90	582.42	116257.54
25-44	132315.0	1078.02	3317.91	23.45	250.68	421.15	835.60	136618.59
45-64	263409.0	1317.57	2963.38	21.98	347.15	582.41	989.52	136999.63
65+	460355.0	1229.90	2664.24	21.98	355.15	587.07	927.58	167924.52

# Counting the total of hospitalizations
# By gender
count_gender = df.groupby(["gender"]).size().reset_index(name='total_hospitalizations')
# By age group
count_age_group = df.groupby(["age_group"]).size().reset_index(name='total_hospitalizations')

# Create the plots
sns.set_theme(style="white", palette="colorblind")

# Bar plot for gender
fig, ax = plt.subplots()

sns.barplot(
    data=count_gender,
    x="gender", y="total_hospitalizations",
    hue='gender',
    ax=ax
)
for container in ax.containers:
    ax.bar_label(container, fontsize=9, color='black', fontweight='bold')
ax.spines[['right', 'top']].set_visible(False)
ax.set_ylabel("Total Hospitalizations")
ax.set_xlabel("")
plt.show()

# Bar plot for age group
fig, ax = plt.subplots()

sns.barplot(
    data=count_age_group,
    x="age_group", y="total_hospitalizations",
    hue='age_group',
    ax=ax
)
for container in ax.containers:
    ax.bar_label(container, fontsize=9, color='black', fontweight='bold')
ax.spines[['right', 'top']].set_visible(False)
ax.set_ylabel("Total Hospitalizations")
ax.set_xlabel("")
plt.show()

(a) ICSAP by Gender (2024)

(b) ICSAP by Age Group (2024)

Figure 4: Total Hospitalizations by Gender and Age Group

# Create the plots
sns.set_theme(style="whitegrid", palette="colorblind")

# Gender
fig, ax = plt.subplots()

sns.boxplot(data=sample_df, x='gender', y='total_paid', hue='gender')
ax.set_ylabel("Total Cost (R$)")
ax.set_xlabel("")
plt.show()

# Gender no outliers
fig, ax = plt.subplots()

sns.boxplot(
    data=sample_df, 
    x='gender', y='total_paid', hue='gender',
    showfliers=False # Hide outliers
) 
ax.set_ylabel("Total Cost (R$)")
ax.set_xlabel("")
plt.show()

# Age group
fig, ax = plt.subplots()

sns.boxplot(data=sample_df, x='age_group', y='total_paid', hue='age_group')
ax.set_ylabel("Total Cost (R$)")
ax.set_xlabel("")
plt.show()

# Age group filtered
fig, ax = plt.subplots()

sns.boxplot(
    data=sample_df,
     x='age_group', y='total_paid', hue='age_group',
     showfliers=False # Hide outliers
)
ax.set_ylabel("Total Cost (R$)")
ax.set_xlabel("")
plt.show()

(a) ICSAP Cost by Gender (2024)

(b) ICSAP Cost by Gender (2024) - Without outs

(c) ICSAP Cost by Age Group (2024)

(d) ICSAP Cost by Age Group (2024) - Without outs

Figure 5: Total Cost by Gender and Age Group

For gender, males have a higher average cost (R$1,205.09) compared to females (R$1,055.84), and also a higher standard deviation (R$2,802.83 vs. R$2,548.52), indicating greater dispersion in male hospitalization expenses. Both genders share the same minimum cost (R$21.98), but males reach a higher maximum (R$167,924.52), suggesting the presence of extreme outliers.

When analyzing age groups, hospitalization costs tend to increase with age. Children aged 0–4 and 5–14 have the lowest average costs , while adults aged 45–64 and seniors 65+ show the highest averages. The standard deviation also grows with age, reflecting more variability in older populations.

We also can have a look in those patterms by macro-regions wiht the metrics agrouped by gender and age group.

# Prepare data
# Metrics agrouoed by gender
df_region_gender = df.groupby(['big_region_name', 'gender']).size().reset_index(name='total_hospitalizations')

df_region_gender_cost = (df.groupby(['big_region_name', 'gender'])
      .agg(total_cost=('total_paid', 'sum'))
      .sort_values(by='big_region_name', ascending=False)
      .reset_index()
)

# Metrics agrouped by age group
df_region_age = df.groupby(['big_region_name', 'age_group']).size().reset_index(name='total_hospitalizations')

df_region_age_cost = (df.groupby(['big_region_name', 'age_group'])
      .agg(total_cost=('total_paid', 'sum'))
      .sort_values(by='big_region_name', ascending=False)
      .reset_index()
)

# Define the order
region_order = (
    df.groupby('big_region_name')['total_paid']
      .sum()
      .sort_values(ascending=False)
      .index
)

# Create plot
sns.set_theme(style="whitegrid", palette="colorblind")
# For gender count
fig, ax = plt.subplots(figsize=(10, 6))

sns.barplot(
    data=df_region_gender,
    x='big_region_name', y='total_hospitalizations',
    hue='gender',
    ax=ax, order=region_order
)
for container in ax.containers:
    ax.bar_label(container, fontsize=9, color='black', fontweight='bold')
ax.spines[['right', 'top', 'left']].set_visible(False)
ax.legend(title=None, loc="upper center", bbox_to_anchor=(0.5, 1.15), ncol=2)
ax.yaxis.set_visible(False)
ax.set_xlabel("")
plt.show()

# For gender cost
fig, ax = plt.subplots(figsize=(10, 6))

sns.barplot(
    data=df_region_gender_cost,
    x='big_region_name', y='total_cost', hue='gender',
    ax=ax, order=region_order, width=0.9
)
for container in ax.containers:
    values = container.datavalues
    labels = [format_brl(v) for v in values]
    ax.bar_label(container, labels=labels, fontsize=9, color='black', fontweight='bold')
ax.spines[['right', 'top', 'left']].set_visible(False)
ax.legend(title=None, loc="upper center", bbox_to_anchor=(0.5, 1.15), ncol=2)
ax.set_ylabel("")
ax.yaxis.set_visible(False)
plt.show()

(a) ICSAP Hospitalizations (2024)

(b) ICSAP Cost (2024)

Figure 6: Total Hospitalizations and Costs by Macro-Regions and Gender

# Define the order
region_order = (
    df.groupby('big_region_name')['total_paid']
      .sum()
      .sort_values(ascending=False)
      .index
)
age_group_order = (
    df.groupby('age_group')['total_paid']
      .sum()
      .sort_values(ascending=False)
      .index
)

# Create plot
sns.set_theme(style="whitegrid", palette="colorblind")

# For age group count
fig, ax = plt.subplots(figsize=(10, 8))

sns.barplot(
    data=df_region_age,
    x='total_hospitalizations', y='big_region_name',
    orient='h', hue='age_group', hue_order=age_group_order,
    ax=ax, order=region_order
)
for container in ax.containers:
    ax.bar_label(container, fontsize=9, color='black', fontweight='bold')
ax.spines[['right', 'top', 'bottom']].set_visible(False)
ax.legend(title=None, loc="upper center", bbox_to_anchor=(0.5, 1.15), ncol=6)
ax.xaxis.set_visible(False)
ax.set_ylabel("")
plt.show()

# For age group cost
fig, ax = plt.subplots(figsize=(10, 8))

sns.barplot(
    data=df_region_age_cost,
    x='total_cost', y='big_region_name',
    orient='h', hue='age_group', hue_order=age_group_order,
    ax=ax, order=region_order
)
for container in ax.containers:
    values = container.datavalues
    labels = [format_brl(v) for v in values]
    ax.bar_label(container, labels=labels, fontsize=9, color='black', fontweight='bold')
ax.spines[['right', 'top', 'bottom']].set_visible(False)
ax.legend(title=None, loc="upper center", bbox_to_anchor=(0.5, 1.15), ncol=6)
ax.xaxis.set_visible(False)
ax.set_ylabel("")
plt.show()

(a) ICSAP Hospitalizations (2024)

(b) ICSAP Cost (2024)

Figure 7: Total Hospitalizations and Costs by Macro-Regions and Age Group

Across Brazil’s major regions, hospitalization patterns reveal consistent gender and age-related trends. The Southeast region recorded the highest number of hospitalizations overall, with male patients slightly outnumbering females. This male predominance was also observed in the South and North, whereas the Northeast and Center-West showed higher hospitalization counts among females.

Older adults (65+) accounted for the largest share of hospitalizations in every region, particularly in the Southeast (196,559) and Northeast (124,967). The 45–64 age group followed closely.

Cost data mirrored these patterns. The Southeast incurred the highest total hospitalization costs for both genders. When broken down by age group, the 65+ population consistently drove the highest costs across all regions, peaking in the Southeast at over R$259 million. The 45–64 group also contributed substantially, reinforcing the financial impact of aging populations on the healthcare system.

ICSAP Disease Categories

Let’s look at the avoidable disease categories, the main categories and their expenditures.

disease_category = (
    df.groupby('disease_category')
      .agg(total_hospitalizations=('disease_category', 'size'),
           total_cost=('total_paid', 'sum'),
           avg_cost=('total_paid', 'mean'),
           avg_stay=('length_stay', 'mean'))
      .sort_values(by='total_hospitalizations', ascending=False)
)
styled_tb = (
    disease_category.style
    .background_gradient(cmap='YlGnBu', high=0.2, subset=['total_cost', 'avg_cost', 'avg_stay'])
    .format({'total_cost': 'R$ {:,.2f}'})
)

styled_tb

	total_hospitalizations	total_cost	avg_cost	avg_stay
disease_category
Infecção no Rim e Trato Urinário	218241	R$ 126,877,175.19	581.362692	5.674319
Doenças Cerebrovasculares	197963	R$ 365,759,200.23	1847.613949	7.290999
Pneumonias bacterianas	168655	R$ 243,172,884.92	1441.836204	6.828223
Diabetes melitus	136172	R$ 151,124,585.96	1109.806612	6.779984
Gastroenterites Infecciosas e complicações	126223	R$ 51,377,873.76	407.040506	3.333283
Infecção da pele e tecido subcutâneo	61046	R$ 52,273,362.37	856.294636	7.341382
Úlcera gastrointestinal	50844	R$ 60,521,054.37	1190.328345	5.599599
Hipertensão	37069	R$ 18,047,013.21	486.849206	4.222855
Insuficiência Cardíaca	26020	R$ 71,171,979.81	2735.279777	8.264374
Doenças preveníveis por imunização e condições sensíveis	23845	R$ 46,316,432.11	1942.395979	13.730887
Doencas pulmonares	22406	R$ 20,922,481.99	933.789252	4.713470
Deficiências Nutricionais	19316	R$ 24,227,126.98	1254.251759	8.666960
Asma	6492	R$ 4,849,799.21	747.042392	3.428990
Infecções de ouvido, nariz e garganta	2239	R$ 670,022.02	299.250567	2.887896
Doenças relacionadas ao Pré-Natal e Parto	1827	R$ 1,143,375.44	625.821259	6.281883
Eplepsias	1485	R$ 1,473,095.85	991.983737	5.622896
Angina	1132	R$ 5,301,099.39	4682.949991	4.873675
Anemia	1121	R$ 446,850.07	398.617368	3.992864
Doença Inflamatória órgãos pélvicos femininos	1081	R$ 480,510.46	444.505513	2.449584

While urinary tract infections top the list in volume, cerebrovascular diseases dominate in total cost, exceeding R$365 million despite fewer admissions. Bacterial pneumonia and diabetes mellitus also represent high-burden conditions, combining frequent hospitalizations with substantial expenditures.

Interestingly, some conditions—such as heart failure and angina—stand out for their elevated cost per hospitalization. Heart failure, with just over 26,000 admissions, incurred more than R$71 million in costs, while angina, with only 1,132 cases, generated over R$5.3 million.

Immunization-preventable conditions still account for over 23,000 hospitalizations and R$46 million in costs — highlighting gaps in preventive coverage.

# Identify the 3 categories with the most hospitalizations
top3_categories = df['disease_category'].value_counts().nlargest(3).index.tolist()

# Filter the data
df_top3 = df[df['disease_category'].isin(top3_categories)].copy()

# Create plot
sns.set_theme(style="whitegrid", palette="colorblind")

# For total cost
c = sns.catplot(
    data=df_top3,
    x='age_group', y='total_paid', hue='age_group',
    col='disease_category',
    kind='box', col_wrap=3, 
    showfliers=False # Hide outliers
)
c.set_axis_labels("", "Total Cost (R$)")
c.set_titles("{col_name}")
plt.show()

# For length stay
s = sns.catplot(
    data=df_top3,
    x='age_group', y='length_stay', hue='age_group',
    col='disease_category',
    kind='box', col_wrap=3,
    showfliers=False # Hide outliers
)
s.set_axis_labels("", "Length of Stay (Days)")
s.set_titles("{col_name}")
plt.show()

(a) ICSAP Total Cost Distribution (2024)

(b) ICSAP Length of Stay Distribution (2024)

Figure 8: Total Cost and Length of Stay Distribution by Age Group

# Total hosp and cost by macro-regions
treemap_data = df.groupby(['big_region_name', 'disease_category']).agg(
    total_hospitalizations=('disease_category', 'size'),
    total_cost=('total_paid', 'sum')
).sort_values('total_hospitalizations', ascending=False).reset_index()

treemap_top5 = treemap_data.groupby('big_region_name').head(5)

# Get Brazil top disease category
top_category_brasil = df['disease_category'].value_counts().idxmax()
br_top_hosp = (df['disease_category'] == top_category_brasil).sum()
br_top_cost = df.loc[df['disease_category'] == top_category_brasil, 'total_paid'].sum()

root_label = f"Brazil - 2024"

# Create plot
# Treemap plot
fig_treemap = px.treemap(
    treemap_top5,
    path=[px.Constant(root_label), 'big_region_name', 'disease_category'],
    values='total_hospitalizations',
    color='total_cost',
    color_continuous_scale='YlGnBu',
    custom_data=['total_cost'],
    labels={'total_cost': 'Total Cost'}
)
default_hover = (
    "<b>%{label}</b><br>"
    "Total Hospitalizations: %{value:,d}<br>"
    "Total Cost: R$ %{customdata[0]:,.2f}<extra></extra>"
)
fig_treemap.update_traces(hovertemplate=default_hover)

labels = list(fig_treemap.data[0]['labels'])
root_idx = labels.index(root_label)  # o índice do nó raiz
hovertemplates = [default_hover] * len(labels)
hovertemplates[root_idx] = (
    f"<b>{root_label}</b><br>"
    f"Brazil Top Disease Category: <b>{top_category_brasil}</b><br>"
    f"Total Hospitalizations: {br_top_hosp:,d}<br>"
    f"Total Cost: R$ {br_top_cost:,.2f}<extra></extra>"
)
fig_treemap.data[0].hovertemplate = hovertemplates
fig_treemap.update_layout(margin=dict(t=50, l=25, r=25, b=25))
fig_treemap.show()

Figure 9: Top 5 ICSAP Disaese Categories by Macro-Regions (2024)

Conclusions

About the dataset:

• Dataset size and structure: The SIH-SUS dataset is very large; therefore, it is more efficient to run queries that aggregate information before importing the data into Power BI.

• Data quality issues:

  • Missing data is minimal, but there are records with zero cost that need filtering for financial analyses.
  • Both cost and length of stay show high variability and skewness, indicating that averages can be influenced by a small subset of complex cases.

About the data:

• Gradual reduction in ICSAP: The proportion of hospitalizations for primary care–sensitive conditions in the SUS decreased between 2014 and 2024 (from 9.2% to 7.8%), suggesting some improvements, though the absolute number of cases remained stable.

• Significant financial impact: Despite the proportional reduction, absolute costs increased, surpassing R$1.2 billion in 2024, indicating that remaining cases are potentially more complex and costly.

• Marked regional differences:

  • The North and Northeast maintain higher ICSAP rates, reflecting possible gaps in primary care access and coverage.
  • The Southeast accounts for the highest absolute costs and hospitalizations due to its larger population.

• Age and gender profiles: Older adults (45–64 and 65+) account for most hospitalizations and costs, reinforcing the impact of population aging.

• Costs and length of stay: Maybe there is no direct relationship between days of hospitalization and cost, as expenses may depend on several factors beyond duration, including the type and complexity of treatment (e.g., surgeries, ICU), medications and supplies, comorbidities and others. Thus, short stays can be expensive, while some long stays may involve relatively low costs.

• Main disease categories:

  • Urinary tract infections lead in admission volume.
  • Cerebrovascular diseases generate the highest total costs.
  • Bacterial pneumonia, diabetes mellitus, and heart failure also have a significant impact.
  • Vaccine-preventable conditions still account for over 23,000 admissions, highlighting gaps in preventive coverage.

References

1.

Pereira, F. J. R., Silva, C. C. da & Lima Neto, E. de A. Condições Sensíveis à Atenção Primária: Uma revisão descritiva dos resultados da produção acadêmica brasileira. Saúde em Debate 38, 331–342 (2014).

2.

Alfradique, M. E. et al. Internações por condições sensíveis à atenção primária: A construção da lista brasileira como ferramenta para medir o desempenho do sistema de saúde (Projeto ICSAP - Brasil). Cadernos de Saúde Pública 25, 1337–1349 (2009).

3.

Ministério da Saúde, B. Portaria Nº 221, de 17 de abril de 2008. Lista Brasileira de internações por Condições Sensíveis à Atenção Primária. (2008).

4.

PROADESS. Internações por Condições Sensíveis à Atenção Primária (ICSAP): Análise descritiva por sexo e idade e diagnósticos principais. (2022).